Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add skills/m4ttstack/fast-browser/fast-browsingnpx skills add m4ttstack/fast-browser --skill fast-browsinggit clone --depth 1 https://github.com/m4ttstack/fast-browserWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00025 | $0.03048 |
| Opus 5 | $0.00013 | $0.01524 |
| Sonnet 5 | $0.00005 | $0.00610 |
| Haiku 4.5 | $0.00003 | $0.00305 |
Grade A, and why
fast-browsing scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured yesterday.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 270 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Fast Browsing
Minimize browser round trips and observation size. Prefer one informed batch over repeated inspect-and-click cycles.
Start with flows
Before macros, run fast-browser flows find --intent "<task>" --origin <origin> --json via the shell and read the candidates array in its JSON
output.
If a candidate has runnable: true, make exactly one browser_run_code_unsafe
call using its invocation field verbatim: invocation.arguments.filename
and invocation.arguments.args, unedited.
Never run a candidate with runnable: false. Its reasons say why:
pending approval: fast-browser flows approve <name> means ask the human to
run that command; contains js step: not replayable in v1 means the flow
needs re-recording. Do not attempt either yourself.
A failed flow-runner call errors with one of two distinct prefixes; check
which one fired before deciding what to do next.
If the call errors with a message starting FLOW_RUNNER_FAILURE: , parse
the JSON payload after that prefix (failedStep, error, url,
stepsCompleted, locatorFallbacks, and on a locator-miss failure,
candidates, showing what the page actually offered at that step). Do not
retry the flow: fall through to macros, then the fast loop below. Never
hand-edit the flow artifact to fix a failure: the next flows compile
sweep reads this same evidence and heals the artifact automatically when
it is unambiguous. If the flow keeps failing, it quarantines on its own;
re-record it instead.
If the call errors with a message starting SIDECAR_LOST: instead, do not
treat it like the case above: it is not a step failure and does not fall
through to macros. See "When a tool call fails with SIDECAR_LOST" below
before doing anything else.
Scripts vs. discrete steps
flows find first stays the rule no matter what runs next. Scripts
(browser_run_code_unsafe) remain right for exploration and one-off tasks;
the fast loop below is built around them.
When the task is a repeatable multi-step journey on a site with no matching
flow, drive it with discrete tool calls (browser_click, browser_type,
browser_navigate, and so on) instead of one script. The flywheel only
compiles discrete-call sessions into replayable steps. A scripted run
compiles to a single opaque js step instead, the same "contains js step:
not replayable in v1" case already covered above, so it never becomes a
runnable candidate on its own. Driving the journey with discrete calls now is
what lets the next flows compile sweep turn it into something a later
session can run directly.
What ships with it
1 file beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- yesterday First seen · 270 lines · 25 tokens per session scan A 21ca3f70ad78
fast-browsing is a skill published in the GitHub repository m4ttstack/fast-browser (0 stars, last pushed 7d ago), licensed MIT. It adds 25 tokens to every session and 3,048 once invoked, about $0.0001 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
webcmd-browser
Use when a live browser task requires Playwright interaction, authenticated handoff, visible UI verification, or ad-hoc page inspection.
moli-webfetch
Fetch, inspect, crawl, and capture live, JavaScript-rendered websites with Moli. Use when Codex needs current web content, web research, fact lookup, link following, a bounded crawl, client-rendered or response-gated content, network diagnostics, or a standalone HTML, Markdown, JSON, semantic-tree, viewport or…
moli-cdp-server
Start Moli's CDP server and connect Playwright, Puppeteer, or raw CDP clients. Use to run a headless-browser CDP endpoint, replace a Chromium process, attach over CDP, enable real layout and screenshot surfaces, or diagnose CDP discovery, connection, and target startup—even when Moli is not named.
surf
Control Chrome browser via CLI for testing, automation, and debugging. Use when the user needs browser automation, screenshots, form filling, page inspection, network/CPU emulation, DevTools streaming, or AI queries via ChatGPT/Gemini/Perplexity/Grok/AI Studio.
deep-x-research
Deep, exhaustive research on a topic across X (Twitter) by driving Grok (x.com/i/grok) through surf. Use when the user wants comprehensive X research on a concept, technique, trend, tool, or creator scene; needs categorized findings with every claim traceable to post URLs; or when a single Grok query is not enough.
surf-codebase
Navigate and modify surf-cli codebase - Chrome extension + native host for AI browser automation. Use for surf-cli code work, architecture questions, implementing browser control/CDP/accessibility/network features.